A Servo Motor Fault Monitoring Method and System Based on Adaptive Compression Transmission

Abstract

This invention discloses a servo motor fault monitoring method and system based on adaptive compression transmission, comprising: periodically collecting operating data of each servo drive through a PLC; performing adaptive incremental compression processing on the operating data to generate compressed operating data blocks and storing them in a buffer; monitoring the compressed operating data blocks in the buffer through the PLC; when the data volume of the compressed operating data blocks reaches a data threshold, packaging the compressed operating data blocks into data frames and uploading them to a host computer; restoring the data frames and extracting feature vectors; and using the feature vectors to perform anomaly detection and health determination of the servo motor.

Description

Technical Field

[0001] This invention relates to the field of servo motor fault monitoring technology, and in particular to a servo motor fault monitoring method and system based on adaptive compressed transmission. Background Technology

[0002] In the context of large-scale industrial production and green manufacturing, lithium battery dismantling / recycling lines have become a crucial link in the new energy industry chain. These lines typically employ industrial automation equipment, using sequential processes such as dismantling, cutting, and sorting to precisely break down battery packs / cells, maximizing the recovery of metallic materials such as aluminum, copper, nickel, and cobalt, while effectively isolating harmful components. Key dismantling and cutting units in these lines generally utilize servo drives to control high-precision cutting tools and robotic arms to meet the requirements of high efficiency, low loss, and safe operation.

[0003] Currently, industrial sites largely rely on time-based preventative maintenance (TBM), but this method of periodic replacement often leads to resource waste or makes it difficult to promptly identify potential equipment problems. Meanwhile, with the acceleration and intelligent upgrading of production lines, servo drives in key units generate a large amount of operational data in real time (including torque, speed, current, position, vibration, etc.). This raw data contains rich information on health and load evolution, but due to the high acquisition rate and massive data volume, traditional data acquisition methods may suffer from insufficient frequency, high network load, large data storage requirements, or heavy processing pressure on the host computer. This is especially true when millisecond-level sampling intervals are required, where the generated data volume is enormous, significantly burdening network bandwidth and storage resources.

[0004] The purpose of this invention is to design a servo motor fault monitoring method and system based on adaptive compressed transmission to address the problems existing in the prior art. Summary of the Invention

[0005] In view of this, the purpose of this invention is to propose a servo motor fault monitoring method and system based on adaptive compressed transmission, which can solve the above-mentioned problems.

[0006] This invention provides a servo motor fault monitoring method based on adaptive compression transmission, comprising:

[0007] The PLC periodically collects the operating data of each servo drive, performs adaptive incremental compression processing on the operating data, generates compressed operating data blocks, and stores them in a buffer.

[0008] The PLC monitors the compressed running data blocks in the buffer. When the amount of data in the compressed running data block reaches the data threshold, the compressed running data block is packaged into a data frame and uploaded to the host computer.

[0009] After restoring the data frame, feature vectors are extracted, and the servo motor is used for anomaly detection and health assessment.

[0010] The present invention also provides a servo motor fault monitoring system based on adaptive compression transmission, which is used to implement a servo motor fault monitoring method based on adaptive compression transmission.

[0011] The beneficial effects of this invention are:

[0012] First, it employs compression algorithms such as process segmentation, adaptive incremental coding, and run-length coding to automatically adjust compression parameters based on the data characteristics of different process stages. This solves the problems of excessive data volume, high pressure on long-term storage and real-time upload, and easy loss of abnormal details caused by millisecond-level high-speed acquisition in industrial settings. It significantly reduces system data bandwidth and storage pressure (compression rate can reach 30%-90%), ensures that critical data details are not lost, and reduces the processing load of embedded platforms such as PLCs, meeting the requirements of high-frequency acquisition and long-cycle monitoring in industrial applications.

[0013] Secondly, based on dynamic thresholds for data volume, sampling quantity, or time, data frames are packaged and efficiently uploaded via TCP / IP protocol, with additional metadata such as device ID, timestamp, and compression parameters. This solves the problems of congestion caused by high-concurrency small packets, difficulty in data synchronization, poor data traceability, and wasted transmission bandwidth in industrial control networks, ensuring real-time transmission.

[0014] Third, by real-time decoding and restoration of compressed data, various statistical and frequency domain features are extracted using a sliding window method, and multidimensional anomaly scores are generated using the Locality Sensitive Hashing (LSH) method. Feature fusion is then performed using a Graph Neural Network (GNN) based on physical / statistical correlation. Finally, the fused features are input into various big data intelligent models for health status identification and fault discrimination. This solves the problems of incomplete data restoration, limited feature representation capabilities, easy missed or misjudged minor anomalies in traditional methods, as well as insufficient expression of interaction relationships between features, difficulty in model generalization, and lack of interpretability. It significantly improves the sensitivity and accuracy of identifying early, minor, and multi-feature collaborative anomalies, reduces false positives and false negatives, and effectively enhances the anomaly discrimination capability and model robustness in high-dimensional and complex scenarios using methods such as LSH and GNN. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings required in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1This is a flowchart of the method in Example 1. Detailed Implementation

[0017] To facilitate understanding by those skilled in the art, the structure of the present invention will now be described in further detail with reference to the accompanying drawings. It should be understood that, unless otherwise specified, the order of the steps mentioned in this embodiment can be adjusted according to actual needs, and they can even be executed simultaneously or partially simultaneously.

[0018] Example 1

[0019] like Figure 1 As shown, this embodiment of the invention provides a servo motor fault monitoring method based on adaptive compression transmission, including:

[0020] S1 collects the operating data of each servo drive through the PLC in a cyclical manner, performs adaptive incremental compression processing on the operating data, generates compressed operating data blocks, and stores them in a buffer.

[0021] S101 uses a PLC to cyclically collect the actual torque, actual speed, position error, and q-axis current of each servo drive at a preset cycle to obtain the raw data sequence;

[0022] In this step, the preset period can be 1ms-10ms. The PLC opens a data buffer in its internal storage area to temporarily store multiple frames of raw data samples acquired at high speed through the fieldbus.

[0023] S102 divides the raw data sequence into start-up segment, cutting segment, and end segment according to the process stage;

[0024] In this step, the PLC divides the data collected in a complete operation (such as a cutting process) into different stages (e.g., no-load / start-up stage, stable cutting stage, exit / end stage) according to preset logic (e.g., based on time, step trigger signal or data characteristics such as torque slope change).

[0025] S103 performs adaptive incremental compression processing on the data of each process stage after segmentation, and outputs the compressed data block of each process segment.

[0026] S1031 converts each segment of running data into an integer according to a preset scaling factor to obtain an integerized data sequence;

[0027] In this step, integerization not only facilitates subsequent encoding and storage, but also reduces extra data space and floating-point operation pressure when combined with compression algorithms.

[0028] S1032 retains the first data point as the base value for the integer data sequence of each process segment, and represents the remaining data points as the difference with the previous data point, forming an incremental coded data sequence.

[0029] In this step, the first integerized data point of each data packet (or each segment) is used as the base value (Base Value). The subsequent data points are stored as the difference between them and the previous data point (Delta Value), that is, Delta[i] = Value[i] - Value[i - 1]. After delta encoding, the Delta values in the sequences under many working conditions become smaller and 0 appears repeatedly, reducing the subsequent information entropy.

[0030] S1033 Compare each difference after delta encoding with the corresponding threshold of the process segment to obtain an adaptive threshold processing data sequence;

[0031] In this step, different change thresholds (Threshold) are set for each data segment. For example, larger thresholds are set in the gentle start and end segments, and smaller thresholds are set in the sharp cutting segment. The PLC calculates each delta value Delta[i]. This further improves the compression ratio and creates conditions for run-length encoding. There are two adaptive threshold processing methods in this step:

[0032] If the absolute value of the difference is less than the corresponding threshold, a special mark is made;

[0033] In this step, if abs(Delta[i]) < Threshold[segment], it is considered that the change is not significant, and this Delta can be represented by a special mark (such as 0 or a mask) to indicate that it is ignored. This is applicable to scenarios where the entire sequence needs to be retained.

[0034] If the absolute value of the difference exceeds the corresponding threshold, a record is made;

[0035] In this step, only when abs(Value[i] - Last_Transmitted_Value) >= Threshold[segment], Value[i] or its corresponding Delta[i] is recorded / transmitted, but this is slightly different from pure delta storage, but more effective. This is applicable to scenarios of event-triggered storage.

[0036] S1034 For the adaptive threshold processing data sequence, use run-length encoding to compress the continuously occurring same values in tuple form to obtain a compressed run data block;

[0037] In this step, the calculated increment sequence (which may contain many zeros after thresholding) is compressed. Consecutive identical increment values ​​(especially zeros) are detected and encoded as (value, repetition count). For example, the sequence [10, 0, 0, 0, 5, 5, -2] might be encoded as [base value, (10, 1), (0, 3), (5, 2), (-2,1)]. Run-Length Encoding (RLE) is particularly suitable for signals with a large number of repetitions (such as zeros), compressing the data volume exponentially.

[0038] S104 packages and stores the compressed operation data blocks of each process segment along with related metadata into a buffer.

[0039] In this step, in industrial scenarios (such as servo driver monitoring of lithium battery dismantling lines), the system needs to collect various operating parameters (torque, speed, current, etc.) at the millisecond level. However, the raw data volume is extremely large, making it difficult to store and transmit in real time using traditional methods, easily causing system bandwidth and storage resource bottlenecks. Industrial processes have obvious process stages (such as start-up, cutting, and termination), and the data patterns and change rates are different at each stage, so a uniform compression strategy often has limited effect. By using "process-adaptive segmentation + targeted compression" throughout the entire process, segmented processing can greatly improve the compression rate and prevent the loss of anomalies or critical information. Data verification shows that combining the above compression processing can reduce the volume of millisecond-level raw data by 30%-90% (depending on the process and operating conditions), significantly reducing communication and storage pressure while ensuring that critical information is not lost. The compressed data facilitates the backend to quickly reconstruct the original time sequence, enabling accurate health status perception, anomaly detection, or fault trend prediction.

[0040] S2 monitors the compressed running data blocks in the buffer through the PLC. When the amount of data in the compressed running data block reaches the data threshold, it packages and encapsulates the compressed running data block into a data frame and uploads it to the host computer.

[0041] Furthermore, the data frame includes: device identifier ID, start timestamp, original number of samples, compression algorithm identifier, compression parameters used, baseline values ​​for each segment, and compressed data payload.

[0042] In this step, the PLC monitors the amount of processed (compressed) data or the cumulative number of original samples / time in the buffer. When a preset threshold is reached (e.g., the number of compressed data bytes, 50-100 original samples, or a time interval of 500ms), a packaging operation is triggered. The PLC packages the compressed data accumulated in the buffer (including the baseline value and the RLE encoding result of the incremental sequence) into a single data frame. This data frame contains necessary metadata (such as device ID, start timestamp, number of original samples, compression algorithm identifier, compression parameters used (such as scaling factor, corresponding threshold), and baseline value) as well as the compressed data payload, facilitating unpacking and restoration by the host computer and subsequent traceability and analysis.

[0043] As a TCP client, the PLC actively establishes a connection with the preset host computer TCP server and sends the packaged compressed data frames to the host computer server in one go via the TCP / IP protocol.

[0044] S3 restores the data frame and extracts feature vectors, then uses these feature vectors to detect anomalies and determine the health of the servo motor.

[0045] S301 acquires data frames and decodes and restores the data frames step by step according to the compression algorithm identifier to obtain the original data sequence.

[0046] S302 performs sliding time window segmentation on the original data sequence, extracts features from the data in each window, and forms a real-time feature vector;

[0047] The feature vector includes: mean, standard deviation, RMS, peak value, kurtosis, skewness, and FFT energy;

[0048] S303 uses Locality Sensitive Hash (LSH) algorithm to perform anomaly detection and health assessment on real-time feature vectors.

[0049] The S3031 acquires baseline data under healthy operating conditions, extracts health feature vectors from the baseline data, and uses the health feature vectors to construct a local sensitive hash index.

[0050] S3032 records the number density of healthy feature vectors falling into each bucket in the Local Sensitive Hash Index, which serves as the distribution benchmark for the health status, thus obtaining the health density;

[0051] S3033 queries the local sensitive hash index for the real-time feature vector to determine the bucket in each local sensitive hash index;

[0052] S3034 calculates the multidimensional anomaly score based on the health density of the bucket into which the real-time feature vector falls, and then fuses the multidimensional anomaly score and the real-time feature vector to obtain the global fused feature vector.

[0053] The health density anomaly score is calculated based on the number of healthy samples falling into the bucket in real-time feature vectors and the maximum number of healthy samples in all buckets. The calculation formula is as follows:

[0054] ,

[0055] in, This indicates the number of healthy samples in the bucket containing the real-time feature vector. Maximum number of healthy samples across all buckets;

[0056] Distance anomaly score is calculated by the average distance between the real-time feature vector and the healthy feature vector within the bucket. The calculation formula is as follows:

[0057] ,

[0058] in, x Represents the real-time feature vector. B This indicates the bucket that x falls into. This represents the i-th eigenvector within the bucket. This indicates the number of healthy feature vectors within the bucket; dist() represents the distance metric.

[0059] The nearest neighbor anomaly score is calculated by counting the number of healthy features found within the K nearest neighbors. The calculation formula is as follows:

[0060]

[0061] in, Indicates the number of adjacent health characteristics. This indicates the maximum number of observable nearest neighbors for a healthy sample.

[0062] In this step, This indicates the number of healthy samples among the K nearest neighbors; fewer healthy samples indicate a greater anomaly. The maximum observable nearest neighbor distance for a healthy sample can be obtained from historical data. Locality-Sensitive Hashing (LSH) is a special type of hashing algorithm that makes similar input data points under a certain "similarity metric" more likely to be hashed into the same bucket, while dissimilar points are hashed into different buckets.

[0063] Furthermore, the global fused feature vector obtained by fusing the multidimensional anomaly score and the real-time feature vector includes:

[0064] Each real-time feature vector and multidimensional anomaly score are used as nodes in a graph network. Connections between nodes are established based on physical relevance, and corresponding weights are assigned to each edge to construct a feature graph network.

[0065] By using a graph neural network, the features of each node are aggregated and updated based on the features of neighboring nodes and edge weights to obtain the final features of all nodes.

[0066] The final features of all nodes are integrated into a global fusion feature vector through pooling operations.

[0067] In this step, data characteristics from the industrial site (such as average torque, q-axis current, etc.) are considered. Features such as anomaly scores are not independent of each other; many features have clear physical or statistical relationships (e.g., torque = current × constant, position error is related to speed fluctuations). Traditional feature concatenation (simply arranging these features in a row and inputting them into the model) ignores these relationships, easily leading to weak model discrimination and poor generalization ability. Setting edge weights through physical correlation allows the model to focus on information flows with causal relationships or strong coupling, improving the fusion effect. Graph neural networks, because they support the propagation and fusion of information between nodes, are more likely to model the influence paths between features compared to conventional MLPs / fully connected networks. This allows them to find the most critical feature collaboration patterns during the learning process, fusing and suppressing noise and redundancy, and amplifying the true signal.

[0068] The S3035 inputs the globally fused feature vector into the fault analysis model and outputs the current potential fault type and its confidence level.

[0069] In this step, the fault analysis model can be constructed using random forest, XGBoost, multilayer perceptron (MLP), deep neural network, LightGBM, etc. It utilizes historically collected data such as actual torque, actual speed, position error, q-axis current to extract the mean, standard deviation, RMS, peak value, kurtosis, skewness, FFT energy, and the calculated health density anomaly score. Distance anomaly score , Number of abnormal neighbors The resulting global fusion feature vector is used to label the corresponding fault type and serves as the training sample for the model.

[0070] Statistical features alone can make it difficult to distinguish some minor anomalies. LSH local density and distance scores can be effectively fused to help identify faults and anomalies. Traditional single-score thresholds are prone to false alarms. The fusion judgment of big data models is more objective. Big data models can accurately determine the fault type and its probability through training with a large amount of data.

[0071] Example 2

[0072] This embodiment provides a servo motor fault monitoring system based on adaptive compressed transmission, which is used to implement a servo motor fault monitoring method based on adaptive compressed transmission.

[0073] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0074] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0075] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0076] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0077] It should be noted that any reference signs placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The words first, second, and third, etc., do not indicate any order. These words can be interpreted as names.

[0078] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0079] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0080] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0081] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

Claims

1. A servo motor fault monitoring method based on adaptive compressed transmission, characterized in that, include: The PLC periodically collects the operating data of each servo drive, performs adaptive incremental compression processing on the operating data, generates compressed operating data blocks, and stores them in a buffer. Specifically: The PLC collects the actual torque, actual speed, position error, and q-axis current of each servo drive in a preset cycle to obtain the raw data sequence. The original data sequence is divided into three stages according to the process: start-up, cutting, and end. The data of each process stage after segmentation are subjected to adaptive incremental compression processing, and the compressed data block of each process stage is output. The compressed operation data blocks of each process segment and related metadata are packaged and stored in the buffer; The PLC monitors the compressed running data blocks in the buffer. When the amount of data in the compressed running data block reaches the data threshold, the compressed running data block is packaged into a data frame and uploaded to the host computer. After restoring the data frame, feature vectors are extracted. These feature vectors are then used for servo motor anomaly detection and health assessment. Specifically: Acquire data frames, and decode and restore the data frames step by step according to the compression algorithm identifier to obtain the original data sequence; The original data sequence is segmented into sliding time windows, and features are extracted from the data in each window to form a real-time feature vector; The Locality Sensitive Hash algorithm is used to perform anomaly detection and health assessment on real-time feature vectors. Specifically: Baseline data of the equipment under healthy operating conditions is collected, health feature vectors of the baseline data are extracted, and local sensitive hash indexes are constructed using the health feature vectors. Record the number density of healthy feature vectors falling into each bucket in the locality-sensitive hash index, and use it as a distribution benchmark for health status to obtain the health density; The real-time feature vector is queried in the Local Sensitive Hash Index to determine the bucket it falls into in each Local Sensitive Hash Index; Based on the health density of the bucket into which the real-time feature vector falls, a multidimensional anomaly score is calculated. The multidimensional anomaly score and the real-time feature vector are then fused to obtain a global fused feature vector. The fused feature vector is input into the fault analysis model, which outputs the current potential fault type and its probability.

2. The servo motor fault monitoring method based on adaptive compression transmission according to claim 1, characterized in that, The step of performing adaptive incremental compression processing on the segmented data of each process stage, and outputting the compressed data block of each process segment includes: Each segment of running data is converted to an integer according to a preset scaling factor to obtain an integer data sequence; For each process segment's integer data sequence, the first data point is retained as the baseline value, and the remaining data points are represented by the difference from the previous data point, forming an incrementally encoded data sequence. Each difference after incremental encoding is compared with the corresponding threshold of the process segment to obtain an adaptive threshold processing data sequence. For adaptive threshold processing data sequences, run-length encoding is used to compress consecutively occurring identical values ​​into tuples, resulting in compressed running data blocks.

3. The servo motor fault monitoring method based on adaptive compression transmission according to claim 2, characterized in that, The process of comparing each difference after incremental encoding with the corresponding threshold of the process segment to obtain the adaptive threshold processing data sequence includes: If the absolute value of the difference is less than the corresponding threshold, it is marked specially; If the absolute value of the difference exceeds the corresponding threshold, it is recorded.

4. The servo motor fault monitoring method based on adaptive compression transmission according to claim 1, characterized in that, The data frame includes: device identifier ID, start timestamp, number of original samples, compression algorithm identifier, compression parameters used, baseline values ​​for each segment, and compressed data payload.

5. The servo motor fault monitoring method based on adaptive compression transmission according to claim 1, characterized in that, The calculation of the multidimensional anomaly score based on the health density of the buckets into which the real-time feature vector falls includes: The health density anomaly score is calculated based on the number of healthy samples falling into the bucket in real-time feature vectors and the maximum number of healthy samples in all buckets. The calculation formula is as follows: ， in, This indicates the number of healthy samples in the bucket containing the real-time feature vector. Maximum number of healthy samples across all buckets; Distance anomaly score is calculated by the average distance between the real-time feature vector and the healthy feature vector within the bucket. The calculation formula is as follows: ， in, x Represents the real-time feature vector. B This indicates the bucket that x falls into. This represents the i-th eigenvector within the bucket. This indicates the number of healthy feature vectors within the bucket; dist() represents the distance metric. The nearest neighbor anomaly score is calculated by counting the number of healthy features found within the K nearest neighbors. The calculation formula is as follows: in, Indicates the number of adjacent health characteristics. This represents the maximum number of observable nearest neighbors for a healthy sample.

6. The servo motor fault monitoring method based on adaptive compression transmission according to claim 1, characterized in that, The process of fusing multidimensional anomaly scores and real-time feature vectors to obtain a global fused feature vector includes: Each real-time feature vector and multidimensional anomaly score are used as nodes in a graph network. Connections between nodes are established based on physical relevance, and corresponding weights are assigned to each edge to construct a feature graph network. By using a graph neural network, the features of each node are aggregated and updated based on the features of neighboring nodes and edge weights to obtain the final features of all nodes. The final features of all nodes are integrated into a global fusion feature vector through pooling operations.

7. A servo motor fault monitoring system based on adaptive compression transmission, characterized in that, This method is used to implement the servo motor fault monitoring method based on adaptive compressed transmission as described in any one of claims 1-6.

Images